Overview

Today we will explore the logic and design of Broockman and Kalla’s 2016 study, “Durably reducing transphobia: A field experiment on door-to-door canvassing”, from the recruitment of subjects for the study to the delivery of their interventions. Then we will explore whether the intervention had any effect on respondents’ feelings toward transgender individuals.

To accomplish this we will:

Summarize the study (5 Minutes)
Set up our work space (2-3 Minutes)
Load a portion of the replication data (1-2 Minutes)
Get a high level overview of the data (5 minutes)
Describe the distribution of covariates in the full dataset (5 minutes)
Examine the difference in covariates between those who did and did not complete the survey (10 minutes)
Examine the difference in covariates between those assigned to each treatment condition in the study. (10 minutes)
Estimate the average treatment effect of the intervention (10 minutes)
Plot the results and comment on the study (10 minutes)
Take the weekly survey (3-5 minutes)

One of these 9 tasks (excluding the weekly) will be randomly selected as the graded question for the lab.

set.seed(20220217)
graded_question <- sample(1:9,size = 1)
paste("Question",graded_question,"is the graded question for this week")

[1] "Question 2 is the graded question for this week"

You will work in your assigned groups. Only one member of each group needs to submit the html file of lab.

This lab must contain the names of the group members in attendance.

If you are attending remotely, you will submit your labs individually.

Here are your assigned groups for the semester.

Goals

Conceptually, this lab will give you lots of practice calculating means and conditional means. We will use these means to

Describe the characteristics of the study’s population. After Question 5, you should be able to describe what a typical registered voter in Miami-Dade County looks like
Explore differences between participants in the study this population Do people who took the baseline survey differ in systematic ways from people who did not (Question 6)
Assess the empirical implications of the identifying assumptions of this design If treatment was randomly assigned, then participants assigned to treatment condition should on average look similar to participants in the control condition (Question 7)
Estimate the average treatment effect over multiple time periods You’ll recall from lecture that the ATE is simply a difference of conditional means:

\[ \begin{align*} E \left[ \frac{\sum_1^m Y_i}{m}-\frac{\sum_{m+1}^N Y_i}{N-m}\right]&=\overbrace{E \left[ \frac{\sum_1^m Y_i}{m}\right]}^{\substack{\text{Average outcome}\\ \text{among treated}\\ \text{units}}} -\overbrace{E \left[\frac{\sum_{m+1}^N Y_i}{N-m}\right]}^{\substack{\text{Average outcome}\\ \text{among control}\\ \text{units}}}\\ &= E [Y_i(1)|D_i=1] -E[Y_i(0)|D_i=0] \end{align*} \]

We’ll see that these quantities can be calculated quickly for multiple variables using the following commands:

We will use the group_by and summarise commands to quickly calculate average values for different groups, and the mutate function to calculate differences between these averages
We will introduce the across() and starts_with() functions to calculate summaries like a mean across multiple variables that start the the same pre-fixes.
We will introduce the pivot_longer(), pivot_wider() and left_join commands to gather, spread, and merge data so that we can calculate the difference of means simply by subtracting means for one group in one column from means for another group in a different column
We will use functions from the ggplot package to graphically display our results.

Please knit this .Rmd file

As with every lab, you should:

Download the file
Save it in your course folder
Update the author: section of the YAML header to include the names of your group members in attendance.
Knit the document
Open the html file in your browser (Easier to read)
Knit the document again after completing a section or chunk (Error checking)
Upload the final lab to Canvas.

1 Summarize the article

What’s Leonhardt’s Basic Claim Broadly the Broockman and Kalla (2016) is interested in questions about persuasion. How do we change peoples minds about controversial political topics. Specifically, they are interested in whether a brief conversation in which participants are given information transgender rights and encouraged to engage in analogic perspective taking can lead to lasting changes in attitudes on these issues.
What are some empirical implications of this claim I would argue the paper draws on research from social psychology that suggests encouraging active processing and perspective taking can lead to durable changes in attitudes

2 Set up your workspace

In the code chunk below, please set up your work space by loading more packages, using the code from slides 9-11 from Tuesday’s class
In the code chunk below, please set up your work space by loading more packages, following the steps outlined in the slides here

# Set working directory
wd <- "." # Change to file path on your computer
setwd(wd)

the_packages <- c(
  ## R Markdown
  "kableExtra","DT",
  ## Tidyverse
  "tidyverse", "lubridate", "forcats", "haven", "labelled",
  ## Extensions for ggplot
  "ggmap","ggrepel", "ggridges", "ggthemes", "ggpubr", 
  "GGally", "scales", "dagitty", "ggdag", "ggforce",
  # Data 
  "COVID19","maps","mapdata","qss","tidycensus", "dataverse",
  # Analysis
  "DeclareDesign", "easystats", "zoo"
)

# Define function to load packages
ipak <- function(pkg){
    new.pkg <- pkg[!(pkg %in% installed.packages()[, "Package"])]
    if (length(new.pkg)) 
        install.packages(new.pkg, dependencies = TRUE)
    sapply(pkg, require, character.only = TRUE)
}

ipak(the_packages)

   kableExtra            DT     tidyverse     lubridate       forcats 
         TRUE          TRUE          TRUE          TRUE          TRUE 
        haven      labelled         ggmap       ggrepel      ggridges 
         TRUE          TRUE          TRUE          TRUE          TRUE 
     ggthemes        ggpubr        GGally        scales       dagitty 
         TRUE          TRUE          TRUE          TRUE          TRUE 
        ggdag       ggforce       COVID19          maps       mapdata 
         TRUE          TRUE          TRUE          TRUE          TRUE 
          qss    tidycensus     dataverse DeclareDesign     easystats 
         TRUE          TRUE          TRUE          TRUE          TRUE 
          zoo 
         TRUE

3 Load the data for today.

Next we’ll load data to explore the phenomenom of Red Covid

3.1 Load Covid-19 data

First we’ll need data on Covid-19 cases and deaths that we’ve worked with throughout the course.

In the chunk below, please write code to load data on Covid-19 in the states using the covid19() function from the COVID19 package

covid <- COVID19::covid19(
  country = "US",
  level = 2,
  verbose = F
)

3.2 Load Election Data

Next need data on the 2020 presidential election. In the code chunk below, write code that will download data presidential elections from 1976 to 2020 from the MIT Election Lab’s dataverse.

The code you’ll need is here

# This joyously stopped working last night...
Sys.setenv("DATAVERSE_SERVER" = "dataverse.harvard.edu")

pres_df <- get_dataframe_by_name(
  "1976-2020-president.tab",
  "doi:10.7910/DVN/42MVDX"
)

Downloading ingested version of data with readr::read_tsv. To download the original version and remove this message, set original = TRUE.

Rows: 4287 Columns: 15
── Column specification ────────────────────────────────────────────────────────
Delimiter: "\t"
chr (6): state, state_po, office, candidate, party_detailed, party_simplified
dbl (7): year, state_fips, state_cen, state_ic, candidatevotes, totalvotes, ...
lgl (2): writein, notes

ℹ Use `spec()` to retrieve the full column specification for this data.
ℹ Specify the column types or set `show_col_types = FALSE` to quiet this message.

# 
# load(url("https://pols1600.paultesta.org/files/data/pres_df.rda"))

Sys.setenv("DATAVERSE_SERVER" = "dataverse.harvard.edu") sets a parameter in your R enivornment that tells the dataverse package to use Harvard’s dataverse
get_dataframe_by_name() downloads the "1976-2020-president.tab" file from the U.S. President 1976–2020 dataverse using its digital object identifier (DOI): doi:10.7910/DVN/42MVDX

4 Describe the structure of each dataset

Specifically what is the unit of observation? Substantively, what does a row covid and pres_df correspond to?

A row in the covid dataset corresponds roughly to a observation of given state (or US Territory) on a given date. For example, the first row in covid descries the state of Covid-19 in the Northern Mariana Islands on March 16, 2020.
A row in the pres_df dataset corresponds to the candidate’s results in a given state in a given presidential election. For example, the first row, describes the number of votes Jimmy Carter got in Alabama in 1976.

You may want to use the code chunk below to get a quick high-level overview of the data

# Take a quick look at your two data sets
head(covid)

        id       date confirmed deaths recovered tests vaccines
1 10b692cc 2020-03-16        NA     NA        NA    NA       NA
2 10b692cc 2020-03-17        NA     NA        NA    NA       NA
3 10b692cc 2020-03-18        NA     NA        NA    NA       NA
4 10b692cc 2020-03-19        NA     NA        NA    NA       NA
5 10b692cc 2020-03-20        NA     NA        NA    NA       NA
6 10b692cc 2020-03-21        NA     NA        NA    NA       NA
  people_vaccinated people_fully_vaccinated hosp icu vent school_closing
1                NA                      NA   NA  NA   NA             NA
2                NA                      NA   NA  NA   NA             NA
3                NA                      NA   NA  NA   NA             NA
4                NA                      NA   NA  NA   NA             NA
5                NA                      NA   NA  NA   NA             NA
6                NA                      NA   NA  NA   NA             NA
  workplace_closing cancel_events gatherings_restrictions transport_closing
1                NA            NA                      NA                NA
2                NA            NA                      NA                NA
3                NA            NA                      NA                NA
4                NA            NA                      NA                NA
5                NA            NA                      NA                NA
6                NA            NA                      NA                NA
  stay_home_restrictions internal_movement_restrictions
1                     NA                             NA
2                     NA                             NA
3                     NA                             NA
4                     NA                             NA
5                     NA                             NA
6                     NA                             NA
  international_movement_restrictions information_campaigns testing_policy
1                                  NA                    NA             NA
2                                  NA                    NA             NA
3                                  NA                    NA             NA
4                                  NA                    NA             NA
5                                  NA                    NA             NA
6                                  NA                    NA             NA
  contact_tracing facial_coverings vaccination_policy elderly_people_protection
1              NA               NA                 NA                        NA
2              NA               NA                 NA                        NA
3              NA               NA                 NA                        NA
4              NA               NA                 NA                        NA
5              NA               NA                 NA                        NA
6              NA               NA                 NA                        NA
  government_response_index stringency_index containment_health_index
1                        NA               NA                       NA
2                        NA               NA                       NA
3                        NA               NA                       NA
4                        NA               NA                       NA
5                        NA               NA                       NA
6                        NA               NA                       NA
  economic_support_index administrative_area_level administrative_area_level_1
1                     NA                         2               United States
2                     NA                         2               United States
3                     NA                         2               United States
4                     NA                         2               United States
5                     NA                         2               United States
6                     NA                         2               United States
  administrative_area_level_2 administrative_area_level_3 latitude longitude
1    Northern Mariana Islands                        <NA> 14.15569  145.2119
2    Northern Mariana Islands                        <NA> 14.15569  145.2119
3    Northern Mariana Islands                        <NA> 14.15569  145.2119
4    Northern Mariana Islands                        <NA> 14.15569  145.2119
5    Northern Mariana Islands                        <NA> 14.15569  145.2119
6    Northern Mariana Islands                        <NA> 14.15569  145.2119
  population iso_alpha_3 iso_alpha_2 iso_numeric iso_currency key_local
1      55144         USA          US         840          USD        69
2      55144         USA          US         840          USD        69
3      55144         USA          US         840          USD        69
4      55144         USA          US         840          USD        69
5      55144         USA          US         840          USD        69
6      55144         USA          US         840          USD        69
  key_google_mobility       key_apple_mobility key_jhu_csse key_nuts key_gadm
1                <NA> Northern Mariana Islands         US69       NA      MNP
2                <NA> Northern Mariana Islands         US69       NA      MNP
3                <NA> Northern Mariana Islands         US69       NA      MNP
4                <NA> Northern Mariana Islands         US69       NA      MNP
5                <NA> Northern Mariana Islands         US69       NA      MNP
6                <NA> Northern Mariana Islands         US69       NA      MNP

head(pres_df)

# A tibble: 6 × 15
   year state   state_po state_fips state_cen state_ic office       candidate   
  <dbl> <chr>   <chr>         <dbl>     <dbl>    <dbl> <chr>        <chr>       
1  1976 ALABAMA AL                1        63       41 US PRESIDENT "CARTER, JI…
2  1976 ALABAMA AL                1        63       41 US PRESIDENT "FORD, GERA…
3  1976 ALABAMA AL                1        63       41 US PRESIDENT "MADDOX, LE…
4  1976 ALABAMA AL                1        63       41 US PRESIDENT "BUBAR, BEN…
5  1976 ALABAMA AL                1        63       41 US PRESIDENT "HALL, GUS" 
6  1976 ALABAMA AL                1        63       41 US PRESIDENT "MACBRIDE, …
# … with 7 more variables: party_detailed <chr>, writein <lgl>,
#   candidatevotes <dbl>, totalvotes <dbl>, version <dbl>, notes <lgl>,
#   party_simplified <chr>

5 Transform data for analysis.

Ok, so our main data set on Covid-19 describes the state of the pandemic in a given state on a given date. To explore the concept of Red Covid, we’ll need to add data on the 2020 election to distinguish Red States from Blue States.

To accomplish this, we’re going to need to:

Recode the Covid-19 data like we’ve done before
Reshape and recode the presidential election data.

5.1 Recode the Covid-19 data

In the chunk below, please recode the covid data to create a covid_us data set, again using code from the slides as your guide, starting here and ending here

# Create a vector containing of US territories
territories <- c(
  "American Samoa",
  "Guam",
  "Northern Mariana Islands",
  "Puerto Rico",
  "Virgin Islands"
  )

# Filter out Territories and create state variable
covid_us <- covid %>%
  filter(!administrative_area_level_2 %in% territories)%>%
  mutate(
    state = administrative_area_level_2
  )

# Calculate new cases, new cases per capita, and 7-day average

covid_us %>%
  dplyr::group_by(state) %>%
  mutate(
    new_cases = confirmed - lag(confirmed),
    new_cases_pc = new_cases / population *100000,
    new_cases_pc_7da = zoo::rollmean(new_cases_pc, 
                                     k = 7, 
                                     align = "right",
                                     fill=NA )
    ) -> covid_us

# Recode facemask policy

covid_us %>%
mutate(
  # Recode facial_coverings to create face_masks
    face_masks = case_when(
      facial_coverings == 0 ~ "No policy",
      abs(facial_coverings) == 1 ~ "Recommended",
      abs(facial_coverings) == 2 ~ "Some requirements",
      abs(facial_coverings) == 3 ~ "Required shared places",
      abs(facial_coverings) == 4 ~ "Required all times",
    ),
    # Turn face_masks into a factor with ordered policy levels
    face_masks = factor(face_masks,
      levels = c("No policy","Recommended",
                 "Some requirements",
                 "Required shared places",
                 "Required all times")
    ) 
    ) -> covid_us

# Create year-month and percent vaccinated variables

covid_us %>%
  mutate(
    year = year(date),
    month = month(date),
    year_month = paste(year, 
                       str_pad(month, width = 2, pad=0), 
                       sep = "-"),
    percent_vaccinated = people_fully_vaccinated/population*100  
    ) -> covid_us

5.2 Create new measures of the 7-day and 14 day averages of new deaths from Covid-19

Using the code from this slide as a guide:

Anywhere you see new_cases write new_deaths
Anywhere you see confirmed write deaths

covid_us %>%
  dplyr::group_by(state) %>%
  mutate(
    new_deaths = deaths - lag(deaths),
    new_deaths_pc = new_deaths / population *100000,
    new_deaths_pc_7da = zoo::rollmean(new_deaths_pc, 
                                     k = 14, 
                                     align = "right",
                                     fill=NA )
    ) -> covid_us

5.3 Reshape and recode the presidential election data.

We want to add election data to our Covid-19 data. To do this, we need to transform our election data, which is structured as candidate-state-election year, into a data set that contains the election results by state for 2020.

Using the code from this slide transform pres_df to create and object called pres2020_df by

Create a copy of the year variable called year_election
Taking the state variable which was ALLCAPS and turning into Title Case
Changing the observations state which are now "District Of Columbia" to "District Of Columbia"
Filtering the data to include only candidates from the Democratic and Republican Parties
Filtering the data to inlcude only the results from the 2020 election.
Selecting the state, year, party_simplified, candidatevotes and totalvotes columns from pres_df
Pivoting the candidatevotes into two new columns with names from the party_simplified column
Creating measures of the Democratic (dem_voteshare)and Republican (rep_voteshare) canditdates’ vote shares in each state by dividing the new DEMOCRAT and REPUBLICAN columns by the values from the totalvotes column
Creating a variable called winner which takes a value of "Trump" if the rep_voteshare variable for a state is greater than the dem_voteshare for a state.
Saving the output of these transformations to an data frame called pres2020_df

Which, I know sounds like a lot. All you need to do is copy and paste the code from this slide.

# Transform Presidential Election data
pres_df %>%
  mutate(
    year_election = year,
    state = str_to_title(state),
    # Fix DC
    state = ifelse(state == "District Of Columbia", "District of Columbia", state)
  ) %>%
  filter(party_simplified %in% c("DEMOCRAT","REPUBLICAN"))%>%
  filter(year == 2020) %>%
  select(state, state_po, year_election, party_simplified, candidatevotes, totalvotes
         ) %>%
  pivot_wider(names_from = party_simplified,
              values_from = candidatevotes) %>%
  mutate(
    dem_voteshare = DEMOCRAT/totalvotes,
    rep_voteshare = REPUBLICAN/totalvotes,
    winner = forcats::fct_rev(factor(ifelse(rep_voteshare > dem_voteshare,"Trump","Biden")))

  ) -> pres2020_df

6 Merge the election data into the Covid-19 data

Now that we’ve got our election data structured as electoral results per state, we can merge this state-level data into our Covid-19 data, using the common variable state in each data frame.

In the code chunk below, use the left_join() command to merge the pres2020_df data into the covid_us data.

Again, you should be able to just copy and paste the code from here

dim(covid_us)

[1] 38174    59

dim(pres2020_df)

[1] 51  9

covid_us <- covid_us %>% left_join(
  pres2020_df,
  by = c("state" = "state")
)
dim(covid_us) # Same number of rows as covid_us w/ 7 additional columns

[1] 38174    67

Ok. That was a lot of code just to get the data set up to do some simple analyses¹.

Let’s start with some simple descriptive statistics.

7 Calculate the average number new Covid-19 deaths in Red and Blue States and median percent of a state’s vaccinated population.

With the covid_us data set:

use the group_by() command to have summarise() calculate values separately by the winner of each state.
use the summarise() command with mean() function to calculate the average number of new deaths (new_deaths) and the average of the 7-day rolling average of new deaths per 100,000 citizens (new_deaths_pc_7da)
use the summarise() command with median() function to calculate median value of the percentage of state’s population that’s vaccinated.
- Remember to tell mean() and median() what to do with NAs using the na.rm argument.

covid_us %>%
  group_by(winner)%>%
  summarise(
    new_deaths = mean(new_deaths, na.rm=T),
    new_deaths_pc_7da = mean(new_deaths_pc_7da, na.rm=T),
    percent_vaccinated = median(percent_vaccinated, na.rm=T)
  )

# A tibble: 2 × 4
  winner new_deaths new_deaths_pc_7da percent_vaccinated
  <fct>       <dbl>             <dbl>              <dbl>
1 Trump        23.8             0.415               43.9
2 Biden        27.1             0.350               54.9

Please interpret the results of your analysis here

8 Calculate the average number new Covid-19 deaths in Red and Blue States using a linear model

Now let’s see how a linear model can be used to estimate conditional means.

Please estimate the following models using the lm() function:

\[ \text{New Deaths} = \beta_0 + \beta_1 \text{Election Winner} \]

\[ \text{7-day average of New Deaths (per 100k)} = \beta_0 + \beta_1 \text{Election Winner} \]

Save the output of lm into objects called m1 and m2.

m1 <- lm(new_deaths ~ winner, covid_us)
m2 <- lm(new_deaths_pc_7da ~ winner, covid_us)

Please interpret the coefficients of the model in terms conditional means you estimated in the previous section

9 Explore the relationships between Republican Vote Share, Vaccination Rates, and Deaths from Covid-19 on September 23, 2021